Louis Lehmann
Lab project · Local AI systems

Hermes

A local-first AI operator running on my Mac mini: part command center, part build partner, part continuity layer for research, websites, browser QA, scheduled checks, and project handoffs.

Mac mini runtimelocal-firstbrowser QAGitHub + Vercelscheduled agents
Architecture visual titled Current Hermes Agentic Setup, showing Hermes Core, operator entry points, memory stores, tools, skills, subagents, runtime ports, roadmap, workspace folders, and launcher scripts.
Hermes / Orion stack visualSnapshot architecture · click to open full size

A reviewable operating layer, not a black-box chatbot.

Hermes sits between intention and execution. It can inspect files, use the browser, run tests, manage GitHub workflows, schedule follow-ups, and keep context across sessions.

Local control center

The system runs from the Mac mini and local Hermes workspace, with click-to-run launchers, dashboard surfaces, and separate project archives.

  • Local files stay inspectable.
  • Workspaces stay separated by project.

Agent workbench

Hermes can use tools rather than just generate text: terminal, files, browser automation, web research, memory, scheduled jobs, and delegated coding agents.

  • Useful for builds, audits, and QA.
  • Verification is part of the loop.

Continuity layer

Long-running projects need memory, handoffs, and restart points. Hermes keeps the operating thread visible so work can resume without re-explaining the whole system.

  • Session search and skills preserve procedure.
  • Roadmaps prevent drift.

Signals in. Tools run. Evidence comes back.

The stack is intentionally practical: everything should produce a visible artifact, a checked route, a committed change, or a concise handoff.

01IntentA request arrives through CLI, browser, Telegram, or a scheduled job.
02ContextHermes loads project files, memories, skills, and prior sessions as needed.
03ActionTools execute: inspect, edit, test, browse, deploy, or delegate.
04VerificationCommands, browser QA, console checks, and live URLs prove what changed.
05ContinuityThe result becomes a repo commit, artifact, skill, note, or next task.

The point is not to make an agent look autonomous. The point is to make useful work easier to start, easier to verify, and easier to continue.

A real system behind the Lab.

Hermes is the operating layer that helps turn small ideas into working pages, dashboards, audits, and recurring checks.

Website shipping

Standalone HTML projects, homepage carousel cards, Lab routes, GitHub commits, Vercel deploy checks, and live visual QA.

Household systems

Family dashboard work, local Mac mini services, calendar integrations, notes, and kitchen-display workflows.

Research and review

Source-checked summaries, repo audits, browser testing, screenshot review, and concise implementation handoffs.

Keep the loop small, visible, and grounded.

Local-first by default

Use local files, local dashboards, and local launchers unless a cloud service clearly earns its place.

Verify before claiming

If work can be checked, run the check. Passing tests, live HTTP responses, browser console output, and screenshots beat vibes.

Separate artifacts from source

Drafts, references, launches, and active repos should have clear homes so the workspace does not become an attic.